Rogue decks

Decks almost nobody plays at tournaments that the trained AI shows are strong. Good picks if you want to surprise an opponent.

Vivi Overflow Stall v4F

A+86.5%vs fieldJP

The rogue Vivi list built to beat the real JP meta: new 170 HP 1st Vivis and a fourth Papa Quits His Job to survive until the opponent decks out.

Read the recipe

Kyoko Hero Cop

S91.7%vs fieldJP

A64.9%head-to-head · #10 RR

Collab every turn for a free cheer, then let 2nd Kyoko hit for 200 with 6 cheers out.

Read the recipe

Kobo-oshi Moona Hiroshima

A+88.2%vs fieldJP

A63.7%head-to-head · #12 RR

An all-Moona deck under Kobo: recycle blue cheers and finish with 130-damage 2nds and special damage.

Read the recipe

Tested against the real JP meta

The tier list plays every deck against one field piloted by Generalist v6. Rogue lists are built to beat the decks people actually bring, so we also test them against the 18 most-played JP decks, each piloted by its own trained AI and weighted by how often its oshi shows up in recent JP tournament lists. Both versions of a deck play the same deals, so the difference between them is measured directly.

Deck	Vs the training field	Vs the real JP meta	Change vs previous version	What changed
Vivi Overflow Stall v4F	86.5% (v1 89.3%)	62.2% (v1 54.6%)	+7.6 ± 1.2	List F ("walls + depth"): +2 1st Vivi hBP09-068 (170 HP), +1 2nd Vivi hBP09-070, +1 Papa Quits His Job, +1 Normal PC, +2 Fugutaro; cuts 2 Archive PC, 2 Oka-nyan, 1 Manager-chan, 1 1st hSD10-008 and 1 Cheer up Curry. Biggest gains: AZKi Frontier 56 → 72, Kronii 24 → 40.
Vivi Overflow Stall v4A	84.1% (no new checkpoint beat v1)	57.5% (v1 52.3%)	+5.2 ± 1.7	List A ("walls"): +2 1st Vivi hBP09-068, +1 2nd Vivi hBP09-070, +1 Papa Quits His Job; cuts 2 Archive PC, 1 Manager-chan and 1 Cheer up Curry. Better than v1, behind v4F. Training didn't improve on the v1 AI for this list, so it plays with the v1 AI.
Riona Burn Plan v2	68.9% (v1 50.5%)	35.8% (v1 25.4%)	+10.4 ± 1.7	Mono-Riona self-mill. Clearly better than v1 and now on the tier list, but still far from a rogue pick.
Moona Full Moon v2	81.6% (v1 83.5%)	48.3% (v1 37.1%)	+11.2 ± 1.7	The JP Bloom Cup winners' list. Better than v1 in all 18 matchups, so the site now uses it.

Meta test: 18 decks × 100 paired deals × 2 versions (3,600 games), share-weighted; ± is one standard error of the difference. The v4F test used 2 deal sets. All results are simulated AI-vs-AI games, not tournament results.

Research notes: Vivi Overflow Stall

Two quick tests from community questions. Neither needed new training: we added a simple rule on top of the trained AIs and replayed the same deals.

Can you beat Vivi by keeping your hand small?

2nd Vivi hBP09-070's Vivacious Vision gets +70 when you hold 7+ cards, and Vivi's forced draws push you toward deck-out. We gave each of the 18 meta AIs a rule: don't take draws that push your hand over the limit, and don't end your main phase over the limit while you can still play cards out. Vivi Overflow Stall v4F, 50 deals per matchup, 2,700 games.

Opponents	Vivi wins	Opponent at 7+ cards when Vivi attacked	Vivi wins by deck-out
No rule	63.3%	80% of attacks	40%
Keep hand at 6 or less	65.2%	76%	25%
Keep hand at 5 or less	62.2%	72%	22%

Playing around doesn't beat Vivi (the differences are within about ±2). Most of the hand growth isn't the opponent's choice: it comes from their draw step and Vivi's forced draws, and their stage fills up, so there's nowhere to put cards.

It changes how Vivi wins (fewer deck-outs, more wins by life), not whether. It helped some matchups (Iroha, FUWAMOCO, Marine, Lamy) and hurt others (Kronii, Moona, Mio-oshi Baelz). The real counter is still a hand reset like Manager-chan or Borderline Evil Robot.

Should Vivi save her SP as a finisher?

Vivi's SP, Nice Vivi! Charming Vivi! Cool Vivi! (3 holo power, once per game, needs a 2nd Vivi in the center), shuffles both hands and all holomem and support cards in both archives back into the decks, then both players draw 7. The opponent is then at 7 cards, so Vivacious Vision hits for 190, but their deck refills. Same setup, 40 deals per matchup.

How Vivi plays	Vivi wins
As trained	63.0%
Use the SP whenever it's legal	62.9%
Save 3 holo power all game for the SP	41.1%
Save power only near the end, then use the SP	62.1%

The SP finisher adds no wins. The trained AI already uses the SP whenever it can, and it is rarely legal: it needs 3 holo power, and the 2-power draw skill Legendary Beauty (V) is the engine of the deck. Holding power for the SP all game costs about 22 points.

Saving power only for the finish tripled how often the SP fired, but mostly in games Vivi was already winning. Vivi loses by dying early, while the opponent's life is still high. The beatdown list, Vivi FLOW GLOW, showed the same thing (45.7% → 42.7%).